Crate burn_dispatch 

Burn multi-backend dispatch.

Available Backends

The dispatch backend supports the following variants, each enabled via cargo features:

Backend	Feature	Description
Cube	cpu, cuda, metal, rocm, vulkan, webgpu, wgpu	Every cubecl runtime. One backend: the features decide which runtimes are compiled in, and a tensor’s device says which one it runs on
Flex	flex	Pure Rust CPU backend using burn-flex
NdArray	ndarray	Pure Rust CPU backend using ndarray (deprecated - use flex)
LibTorch	tch	Libtorch backend via tch (deprecated - use a CubeCL backend)
Autodiff	autodiff	Autodiff-enabled backend (used in combination with any of the backends above)

Note: The features can be combined freely. The cubecl-backed ones all select the same backend, so they share the one DispatchDevice::Cube variant — enabling several compiles several runtimes in, and the device a tensor carries is what picks between them.

Re-exports

pub use backend::*;

pub use device::*;

pub use tensor::*;

Modules

backend

Dispatch backend module.

backends

Backends and devices used.

device

Dispatch device module.

devices

Backend devices.

tensor

Dispatch tensor module.